I've been running Lindy through its paces for the last month, trying to automate parts of our support and internal ticketing. The marketing says "AI-powered," but what I'm seeing under the hood looks a lot like the deterministic automation we've had for years, just with a nicer wrapper.
I set up a "Smart Responder" for common technical queries. The moment you step outside the obvious path, it falls apart. It's not reasoning; it's matching keywords and executing pre-defined flows. I tested it with slight variations in user requests that a human (or a decent LLM) would understand as identical.
Example: I trained it on "Reset my password" and "I can't log in, need a new password." Works fine. Then I throw it a curveball: "My credentials seem to be invalid, can you help me regain access?" It either defaults to a generic fallback or picks the wrong workflow entirely. That's not AI, that's brittle pattern matching.
Here's the core of my issue:
* **Lack of true comprehension:** It doesn't parse intent from context it hasn't been explicitly trained on. It's mapping input A to action B.
* **No dynamic adaptation:** If the process changes, you're back in the builder re-drawing the flowchart. A real AI agent should be able to adjust its steps given a new high-level instruction.
* **The "AI" label is on the builder, not the bot:** The assistive features for *creating* the bots might use some LLM, but the running bot itself behaves like a classic rules engine.
I want to be proven wrong. Has anyone pushed Lindy into a scenario where it genuinely had to *figure something out* on the fly, not just follow the yellow brick road you laid down for it? Show me the logs or the config that demonstrates a non-deterministic, reasoning step.
-- bb
-- bb
You're absolutely right about the intent parsing being the weak spot. I tried using it for lead qualification from website chat, and hit the same wall. It could handle "I want a demo" but not "Can someone walk me through the pricing and features?"
The thing is, for simple, high-volume tasks like password resets, that deterministic flow is actually reliable and fast. I'd trust it over a creative but unpredictable LLM for that specific job. But they really shouldn't market it as general "AI." It's advanced workflow automation, not intelligence.
Have you found any tool that does the true comprehension part well for support? I'm still hunting.
one stack at a time